Margin and Consistency Supervision for Calibrated and Robust Vision Models

2026-03-06

AI Summary

Deep visual classifiers often exhibit poor calibration and weak robustness under distribution shifts, despite achieving high accuracy on clean data. To address this, this work proposes MaCS, a novel framework that unifies maximal class separation and local prediction stability within a single theoretical formulation. By introducing a hinge-squared margin penalty and a KL-divergence consistency regularizer in logit space, MaCS jointly optimizes calibration and robustness without requiring additional data or architectural modifications. Theoretical analysis yields provable bounds on generalization error and robustness radius. Extensive experiments demonstrate that MaCS significantly reduces expected calibration error (ECE) and negative log-likelihood (NLL), enhances robustness to common corruptions, and maintains or even improves top-1 accuracy across diverse image classification benchmarks and backbone architectures.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Adversarial Attacks & Robustness
• Natural Language Processing: Safety and Robustness

Application Category

• Security and Privacy: Security and privacy of machine learning and AI applications
• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Deep vision classifiers often achieve high accuracy while remaining poorly calibrated and fragile under small distribution shifts. We present Margin and Consistency Supervision (MaCS), a simple, architecture-agnostic regularization framework that jointly enforces logit-space separation and local prediction stability. MaCS augments cross-entropy with (i) a hinge-squared margin penalty that enforces a target logit gap between the correct class and the strongest competitor, and (ii) a consistency regularizer that minimizes the KL divergence between predictions on clean inputs and mildly perturbed views. We provide a unifying theoretical analysis showing that increasing classification margin while reducing local sensitivity formalized via a Lipschitz-type stability proxy yields improved generalization guarantees and a provable robustness radius bound scaling with the margin-to-sensitivity ratio. Across several image classification benchmarks and several backbones spanning CNNs and Vision Transformers, MaCS consistently improves calibration (lower ECE and NLL) and robustness to common corruptions while preserving or improving top-1 accuracy. Our approach requires no additional data, no architectural changes, and negligible inference overhead, making it an effective drop-in replacement for standard training objectives.

Problem

• model calibration
• distribution shift
• robustness
• vision models
• classification margin

Innovation

• margin regularization
• consistency regularization
• model calibration
• robustness
• Lipschitz stability